Nguyen et al. (2026) PERSIANN-Unet: A Global Deep Learning Framework for Near-Real-Time Precipitation Estimation Using Infrared Data

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Journal of Hydrometeorology
• Year: 2026
• Date: 2026-03-23
• Authors: Phu Nguyen, Vu Dao, Tu Thanh Ung, Claudia Jimenez Arellano, Kuolin Hsu, Soroosh Sorooshian, Amir AghaKouchak, George J. Huffman, F. Martin Ralph
• DOI: 10.1175/jhm-d-25-0162.1

Research Groups

Not explicitly stated in the provided abstract.

Short Summary

This study introduces PERSIANN-Unet (PUnet), a new quasi-global, high-resolution, near-real-time precipitation algorithm leveraging infrared (IR) data and a UNet architecture. PUnet provides half-hourly, 0.04° precipitation estimates, closely matching its training target (IMERG V07 Final) globally and demonstrating good performance against Stage IV over the Continental United States (CONUS).

Objective

• To develop and evaluate PERSIANN-Unet (PUnet or PERSIANN V3), a quasi-global, deep learning-based algorithm that combines infrared (IR) data and monthly climatology using a UNet architecture to produce half-hourly precipitation estimates at 0.04° resolution, with a focus on an IR-based framework not reliant on passive microwave (PMW) availability.

Study Configuration

• Spatial Scale: Quasi-global, covering 60°N–60°S, with a spatial resolution of 0.04°.
• Temporal Scale: Half-hourly precipitation estimates, designed for near-real-time application. Evaluation period: 2022–2023. Training period: 2016–2021.

Methodology and Data

• Models used: Deep learning, specifically convolutional neural networks (CNNs) with a UNet architecture.
• Data sources:
  • Input: Geosynchronous thermal infrared (IR) data and monthly climatology.
  • Training Target: IMERG V07 Final (an integrated PMW-IR-gauge precipitation product).
  • Evaluation References: HE, IMERG, PDIR-Now, and Stage IV (over CONUS).

Main Results

• PERSIANN-Unet (PUnet) closely matches its training target, IMERG V07 Final, at the global scale.
• PUnet's performance was further evaluated against Stage IV as a reference over the Continental United States (CONUS).
• By operating on a single global image, PUnet avoids tile partitioning and blending steps, which reduces edge discontinuities and produces more spatially consistent precipitation fields across hemispheres.

Contributions

• Introduction of PERSIANN-Unet (PERSIANN V3), a novel quasi-global, high-resolution, near-real-time precipitation product.
• Development of an infrared (IR)-based precipitation estimation framework that is not reliant on passive microwave (PMW) availability, addressing limitations of PMW sensors.
• Implementation of a single global image processing approach that eliminates tile partitioning and blending, leading to reduced edge discontinuities and improved spatial consistency of precipitation fields.

Funding

Not explicitly stated in the provided abstract.

Citation

@article{Nguyen2026PERSIANNUnet,
  author = {Nguyen, Phu and Dao, Vu and Ung, Tu Thanh and Arellano, Claudia Jimenez and Hsu, Kuolin and Sorooshian, Soroosh and AghaKouchak, Amir and Huffman, George J. and Ralph, F. Martin},
  title = {PERSIANN-Unet: A Global Deep Learning Framework for Near-Real-Time Precipitation Estimation Using Infrared Data},
  journal = {Journal of Hydrometeorology},
  year = {2026},
  doi = {10.1175/jhm-d-25-0162.1},
  url = {https://doi.org/10.1175/jhm-d-25-0162.1}
}